[ltp] Re: Re: Stress testing for undervolting

[Laurent Gilson]

Hello,


>> Then 2 hours torture test with all the things i wrote in the first mail
>> (mprime, seeks, WLAN, changing the frequency via script every 30 sec).  
>> No
>> crashs, no error in mprime. But it still didnīt "feel" safe so i added
>> another step to each voltage.
>
> Do you still have this script?

Nope. it was one file with lines like:

echo "600000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
echo "600000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
sleep 10
umount /dvd
mount /dev/hdc /dvd
find /dvd > /dev/null & #scannes the whole DVD-fs in background
sleep 10
echo "1200000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
echo "1200000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
sleep 10
#start the seek testing prog in background

sleep 10
echo "1000000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
echo "1000000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
killall find
sleep 10
umount /dvd
dd if=/dev/hdc of=/dev/null # copies the whole dvd as fast as possible
sleep 20
echo "800000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
echo "800000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq

and so on.

> In your first post you mentioned that you unplugged/replugged the AC and  
> the
> battery.  Did you do that physically/manually?

Manually. tp_smapi support for loading/unload did not work yet for me.

>> After 3 hours surfing/office my CPU is at 42-45°C, w/o any fan running.  
>> I
>> have 3:45 runtime from a 40Wh battery (with WLAN switched on but  
>> bluetooth
>> off) (IBM promises 3:00 for a 47Wh... go figure). I didnīt bother to  
>> check
>> if the last "safety" step was really necessary.
>
> That's nice.  I'm curious.  It sounds as though you're about as careful  
> with
> under-volting as I would like to be.  Did you try to dig a little more  
> power conservation by messing with the fan speeds/triggers?

no, the fan still works with the default values, i did not mess with these  
sensitive parts.

What i did:
- compress my debian with squashfs, load that image at startup into RAM,  
switch off the HD altogether.
- undervolt the CPU
- conservative speedstep-gov.
- underclock the gfx-card (120/120 MHz in 24 bit colors, 70/70 MHz in 16  
bit colors. 16 bit colors saves about 1.5-2 watt => 15-20% more runtime.  
only used it desperate cases)
- use "iwconfig eth0 power period 1000000" and "iwpriv eth0 set_power 5".  
And txpower 10dbm (donīt know if the txpower is really set. It makes 0  
difference in runtime ... ?)
- disabled GLX (i usually donīt need 3D-acc. and it saves 2 watt. GLX  
keeps the CPU in a higher C-State)

> It seems to me that if temperatures are really
> lower, then fan usage should drop without any further tinkering.

The hard point for me is the gfx-card. Itīs the only thing that can still  
trigger the fan in office use. Since i have a radeon 7500 (unsupported by  
ati under linux) i cannot use powerplay. powerplay should slove that  
problem easily. And that chip takes 15-20 mins to cool down from 49°C  
(first trigger point) to < 40°C (fan stops).

> Also, was mprime you sole indicator that your voltage might be too low?

Yes. mprime, running in CPU-intensive stetting.

cu